AAAI 2023 Papers

The creation of a parameterized stylized character involves careful selection of numerous parameters, also known as the "avatar vectors" that can be interpreted by the avatar engine. Existing unsupervised avatar vector estimation methods that auto-create avatars for users, however, often fail to work because of the domain gap between realistic faces and stylized avatar images. To this end, we propose SwiftAvatar, a novel avatar auto-creation framework that is evidently superior to previous works. SwiftAvatar introduces dual-domain generators to create pairs of realistic faces and avatar images using shared latent codes. The latent codes can then be bridged with the avatar vectors as pairs, by performing GAN inversion on the avatar images rendered from the engine using avatar vectors. Through this way, we are able to synthesize paired data in high-quality as many as possible, consisting of avatar vectors and their corresponding realistic faces. We also propose semantic augmentation to improve the diversity of synthesis. Finally, a light-weight avatar vector estimator is trained on the synthetic pairs to implement efficient auto-creation. Our experiments demonstrate the effectiveness and efficiency of SwiftAvatar on two different avatar engines. The superiority and advantageous flexibility of SwiftAvatar are also verified in both subjective and objective evaluations.

Machine learning systems are often deployed for making critical decisions like credit lending, hiring, etc. While making decisions, such systems often encode the user's demographic information (like gender, age) in their intermediate representations. This can lead to decisions that are biased towards specific demographics. Prior work has focused on debiasing intermediate representations to ensure fair decisions. However, these approaches fail to remain fair with changes in the task or demographic distribution. To ensure fairness in the wild, it is important for a system to adapt to such changes as it accesses new data in an incremental fashion. In this work, we propose to address this issue by introducing the problem of learning fair representations in an incremental learning setting. To this end, we present Fairness-aware Incremental Representation Learning (FaIRL), a representation learning system that can sustain fairness while incrementally learning new tasks. FaIRL is able to achieve fairness and learn new tasks by controlling the rate-distortion function of the learned representations. Our empirical evaluations show that FaIRL is able to make fair decisions while achieving high performance on the target task, outperforming several baselines.

Significant progress has been made in representation learning, especially with recent success on self-supervised contrastive learning. However, for time series with less intuitive or semantic meaning, sampling bias may be inevitably encountered in unsupervised approaches. Although supervised contrastive learning has shown superior performance by leveraging label information, it may also suffer from class collapse. In this study, we consider a realistic scenario in industry with limited annotation information available. A supervised contrastive framework is developed for high-frequency time series representation and classification, wherein a novel variant of supervised contrastive loss is proposed to include multiple augmentations while induce spread within each class. Experiments on four mainstream public datasets as well as a series of sensitivity and ablation analyses demonstrate that the learned representations are effective and robust compared with the direct supervised learning and self-supervised learning, notably under the minimal few-shot situation.

Most existing methods realize 3D instance segmentation by extending those models used for 3D object detection or 3D semantic segmentation. However, these non-straightforward methods suffer from two drawbacks: 1) Imprecise bounding boxes or unsatisfactory semantic predictions limit the performance of the overall 3D instance segmentation framework. 2) Existing method requires a time-consuming intermediate step of aggregation. To address these issues, this paper proposes a novel end-to-end 3D instance segmentation method based on Superpoint Transformer, named as SPFormer. It groups potential features from point clouds into superpoints, and directly predicts instances through query vectors without relying on the results of object detection or semantic segmentation. The key step in this framework is a novel query decoder with transformers that can capture the instance information through the superpoint cross-attention mechanism and generate the superpoint masks of the instances. Through bipartite matching based on superpoint masks, SPFormer can implement the network training without the intermediate aggregation step, which accelerates the network. Extensive experiments on ScanNetv2 and S3DIS benchmarks verify that our method is concise yet efficient. Notably, SPFormer exceeds compared state-of-the-art methods by 4.3% on ScanNetv2 hidden test set in terms of mAP and keeps fast inference speed (247ms per frame) simultaneously. Code is available at https://github.com/sunjiahao1999/SPFormer.

Automatic segmentation of left ventricular endocardium in echocardiography videos is critical for assessing various cardiac functions and improving the diagnosis of cardiac diseases. It is yet a challenging task due to heavy speckle noise, significant shape variability of cardiac structure, and limited labeled data. Particularly, the real-time demand in clinical practice makes this task even harder. In this paper, we propose a novel proxy- and kernel-based semi-supervised segmentation network (PKEcho-Net) to comprehensively address these challenges. We first propose a multi-scale region proxy (MRP) mechanism to model the region-wise contexts, in which a learnable region proxy with an arbitrary shape is developed in each layer of the encoder, allowing the network to identify homogeneous semantics and hence alleviate the influence of speckle noise on segmentation. To sufficiently and efficiently exploit temporal consistency, different from traditional methods which only utilize the temporal contexts of two neighboring frames via feature warping or self-attention mechanism, we formulate the semi-supervised segmentation with a group of learnable kernels, which can naturally and uniformly encode the appearances of left ventricular endocardium, as well as extracting the inter-frame contexts across the whole video to resist the fast shape variability of cardiac structures. Extensive experiments have been conducted on two famous public echocardiography video datasets, EchoNet-Dynamic and CAMUS. Our model achieves the best performance-efficiency trade-off when compared with other state-of-the-art approaches, attaining comparative accuracy with a much faster speed. The code is available at https://github.com/JingyinLin/PKEcho-Net.

Many natural language processing models are perceived to be fragile on adversarial attacks. Recent work on adversarial attack has demonstrated a high success rate on sentiment analysis as well as classification models. However, attacks to summarization models have not been well studied. Summarization tasks are rarely influenced by word substitution, since advanced abstractive summary models utilize sentence level information. In this paper, we propose a paraphrasing-based attack method to attack summarization models. We first rank the sentences in the document according to their impacts to summarization. Then, we apply paraphrasing procedure to generate adversarial samples. Finally, we test our algorithm on benchmarks datasets against others methods. Our approach achieved the highest success rate and the lowest sentence substitution rate. In addition, the adversarial samples have high semantic similarity with the original sentences.

A primary objective of news articles is to establish the factual record for an event, frequently achieved by conveying both the details of the specified event (i.e., the 5 Ws; Who, What, Where, When and Why regarding the event) and how people reacted to it (i.e., reported statements). However, existing work on news summarization almost exclusively focuses on the event details. In this work, we propose the novel task of summarizing the reactions of different speakers, as expressed by their reported statements, to a given event. To this end, we create a new multi-document summarization benchmark, SumREN, comprising 745 summaries of reported statements from various public figures obtained from 633 news articles discussing 132 events. We propose an automatic silver-training data generation approach for our task, which helps smaller models like BART achieve GPT-3 level performance on this task. Finally, we introduce a pipeline-based framework for summarizing reported speech, which we empirically show to generate summaries that are more abstractive and factual than baseline query-focused summarization approaches.

The Sudoku Assistant app is an AI assistant that uses a combination of machine learning and constraint programming techniques, to interpret and explain a pen-and-paper Sudoku scanned with a smartphone. Although the demo is about Sudoku, the underlying techniques are equally applicable to other constraint solving problems like timetabling, scheduling, and vehicle routing.

Despite the great achievements of Graph Neural Networks (GNNs) in graph learning, conventional GNNs struggle to break through the upper limit of the expressiveness of first-order Weisfeiler-Leman graph isomorphism test algorithm (1-WL) due to the consistency of the propagation paradigm of GNNs with the 1-WL.Based on the fact that it is easier to distinguish the original graph through subgraphs, we propose a novel framework neural network framework called Substructure Aware Graph Neural Networks (SAGNN) to address these issues. We first propose a Cut subgraph which can be obtained from the original graph by continuously and selectively removing edges. Then we extend the random walk encoding paradigm to the return probability of the rooted node on the subgraph to capture the structural information and use it as a node feature to improve the expressiveness of GNNs. We theoretically prove that our framework is more powerful than 1-WL, and is superior in structure perception. Our extensive experiments demonstrate the effectiveness of our framework, achieving state-of-the-art performance on a variety of well-proven graph tasks, and GNNs equipped with our framework perform flawlessly even in 3-WL failed graphs. Specifically, our framework achieves a maximum performance improvement of 83% compared to the base models and 32% compared to the previous state-of-the-art methods.

Exploration under sparse rewards is a key challenge for multi-agent reinforcement learning problems. One possible solution to this issue is to exploit inherent task structures for an acceleration of exploration. In this paper, we present a novel exploration approach, which encodes a special structural prior on the reward function into exploration, for sparse-reward multi-agent tasks. Specifically, a novel entropic exploration objective which encodes the structural prior is proposed to accelerate the discovery of rewards. By maximizing the lower bound of this objective, we then propose an algorithm with moderate computational cost, which can be applied to practical tasks. Under the sparse-reward setting, we show that the proposed algorithm significantly outperforms the state-of-the-art algorithms in the multiple-particle environment, the Google Research Football and StarCraft II micromanagement tasks. To the best of our knowledge, on some hard tasks (such as 27m_vs_30m}) which have relatively larger number of agents and need non-trivial strategies to defeat enemies, our method is the first to learn winning strategies under the sparse-reward setting.

Submodular maximization arises in many applications, and has attracted a lot of research attentions from various areas such as artificial intelligence, finance and operations research. Previous studies mainly consider only one kind of constraint, while many real-world problems often involve several constraints. In this paper, we consider the problem of submodular maximization under the intersection of two commonly used constraints, i.e., k-matroid constraint and m-knapsack constraint, and propose a new algorithm SPROUT by incorporating partial enumeration into the simultaneous greedy framework. We prove that SPROUT can achieve a polynomial-time approximation guarantee better than the state-of-the-art algorithms. Then, we introduce the random enumeration and smooth techniques into SPROUT to improve its efficiency, resulting in the SPROUT++ algorithm, which can keep a similar approximation guarantee. Experiments on the applications of movie recommendation and weighted max-cut demonstrate the superiority of SPROUT++ in practice.

Different people speak with diverse personalized speaking styles. Although existing one-shot talking head methods have made significant progress in lip sync, natural facial expressions, and stable head motions, they still cannot generate diverse speaking styles in the final talking head videos. To tackle this problem, we propose a one-shot style-controllable talking face generation framework. In a nutshell, we aim to attain a speaking style from an arbitrary reference speaking video and then drive the one-shot portrait to speak with the reference speaking style and another piece of audio. Specifically, we first develop a style encoder to extract dynamic facial motion patterns of a style reference video and then encode them into a style code. Afterward, we introduce a style-controllable decoder to synthesize stylized facial animations from the speech content and style code. In order to integrate the reference speaking style into generated videos, we design a style-aware adaptive transformer, which enables the encoded style code to adjust the weights of the feed-forward layers accordingly. Thanks to the style-aware adaptation mechanism, the reference speaking style can be better embedded into synthesized videos during decoding. Extensive experiments demonstrate that our method is capable of generating talking head videos with diverse speaking styles from only one portrait image and an audio clip while achieving authentic visual effects. Project Page: https://github.com/FuxiVirtualHuman/styletalk.

Previous remote sensing recognition approaches predominantly perform well on the training-testing dataset. However, due to large style discrepancies not only among multidomain datasets but also within a single domain, they suffer from obvious performance degradation when applied to unseen domains. In this paper, we propose a style-content metric learning framework to address the generalizable remote sensing object recognition issue. Specifically, we firstly design an inter-class dispersion metric to encourage the model to make decision based on content rather than the style, which is achieved by dispersing predictions generated from the contents of both positive sample and negative sample and the style of input image. Secondly, we propose an intra-class compactness metric to force the model to be less style-biased by compacting classifier's predictions from the content of input image and the styles of positive sample and negative sample. Lastly, we design an intra-class interaction metric to improve model's recognition accuracy by pulling in classifier's predictions obtained from the input image and positive sample. Extensive experiments on four datasets show that our style-content metric learning achieves superior generalization performance against the state-of-the-art competitors. Code and model are available at: https://github.com/wdzhao123/TSCM.

Inference-time adaptation methods for semantic parsing are useful for leveraging examples from newly-observed domains without repeated fine-tuning. Existing approaches typically bias the decoder by simply concatenating input-output example pairs (cases) from the new domain at the encoder’s input in a Seq-to-Seq model. Such methods cannot adequately leverage the structure of logical forms in the case examples. We propose StructCBR, a structured case-based reasoning approach, which leverages subtree-level similarity between logical forms of cases and candidate outputs, resulting in better decoder decisions. For the task of adapting Text-to-SQL models to unseen schemas, we show that exploiting case examples in a structured manner via StructCBR offers consistent performance improvements over prior inference-time adaptation methods across five different databases. To the best of our knowledge, we are the first to attempt inference-time adaptation of Text-to-SQL models, and harness trainable structured similarity between subqueries.

Doubly stochastic matrix plays an essential role in several areas such as statistics and machine learning. In this paper we consider the optimal approximation of a square matrix in the set of doubly stochastic matrices. A structured BFGS method is proposed to solve the dual of the primal problem. The resulting algorithm builds curvature information into the diagonal components of the true Hessian, so that it takes only additional linear cost to obtain the descent direction based on the gradient information without having to explicitly store the inverse Hessian approximation. The cost is substantially fewer than quadratic complexity of the classical BFGS algorithm. Meanwhile, a Newton-based line search method is presented for finding a suitable step size, which in practice uses the existing knowledge and takes only one iteration. The global convergence of our algorithm is established. We verify the advantages of our approach on both synthetic data and real data sets. The experimental results demonstrate that our algorithm outperforms the state-of-the-art solvers and enjoys outstanding scalability.

Real depth super-resolution (DSR), unlike synthetic settings, is a challenging task due to the structural distortion and the edge noise caused by the natural degradation in real-world low-resolution (LR) depth maps. These defeats result in significant structure inconsistency between the depth map and the RGB guidance, which potentially confuses the RGB-structure guidance and thereby degrades the DSR quality. In this paper, we propose a novel structure flow-guided DSR framework, where a cross-modality flow map is learned to guide the RGB-structure information transferring for precise depth upsampling. Specifically, our framework consists of a cross-modality flow-guided upsampling network (CFUNet) and a flow-enhanced pyramid edge attention network (PEANet). CFUNet contains a trilateral self-attention module combining both the geometric and semantic correlations for reliable cross-modality flow learning. Then, the learned flow maps are combined with the grid-sampling mechanism for coarse high-resolution (HR) depth prediction. PEANet targets at integrating the learned flow map as the edge attention into a pyramid network to hierarchically learn the edge-focused guidance feature for depth edge refinement. Extensive experiments on real and synthetic DSR datasets verify that our approach achieves excellent performance compared to state-of-the-art methods. Our code is available at: https://github.com/Yuanjiayii/DSR-SFG.

Recommender systems now consume large-scale data and play a significant role in improving user experience. Graph Neural Networks (GNNs) have emerged as one of the most effective recommender system models because they model the rich relational information. The ever-growing volume of data can make training GNNs prohibitively expensive. To address this, previous attempts propose to train the GNN models incrementally as new data blocks arrive. Feature and structure knowledge distillation techniques have been explored to allow the GNN model to train in a fast incremental fashion while alleviating the catastrophic forgetting problem. However, preserving the same amount of the historical information for all users is sub-optimal since it fails to take into account the dynamics of each user's change of preferences. For the users whose interests shift substantially, retaining too much of the old knowledge can overly constrain the model, preventing it from quickly adapting to the users’ novel interests. In contrast, for users who have static preferences, model performance can benefit greatly from preserving as much of the user's long-term preferences as possible. In this work, we propose a novel training strategy that adaptively learns personalized imitation weights for each user to balance the contribution from the recent data and the amount of knowledge to be distilled from previous time periods. We demonstrate the effectiveness of learning imitation weights via a comparison on five diverse datasets for three state-of-art structure distillation based recommender systems. The performance shows consistent improvement over competitive incremental learning techniques.

Numeric planning is known to be undecidable even under severe restrictions. Prior work has investigated the decidability boundaries by restricting the expressiveness of the planning formalism in terms of the numeric functions allowed in conditions and effects. We study a well-known restricted form of Hoffmann's simple numeric planning, which is undecidable. We analyze the complexity by imposing restrictions on the causal structure, exploiting a novel method for bounding variable domain sizes. First, we show that plan existence for tasks where all numeric variables are root nodes in the causal graph is in PSPACE. Second, we show that for tasks with only numeric leaf variables the problem is decidable, and that it is in PSPACE if the propositional state space has a fixed size. Our work lays a strong foundation for future investigations of structurally more complex tasks. From a practical perspective, our method allows to employ heuristics and methods that are geared towards finite variable domains (such as pattern database heuristics or decoupled search) to solve non-trivial families of numeric planning problems.

Stroke extraction of Chinese characters plays an important role in the field of character recognition and generation. The most existing character stroke extraction methods focus on image morphological features. These methods usually lead to errors of cross strokes extraction and stroke matching due to rarely using stroke semantics and prior information. In this paper, we propose a deep learning-based character stroke extraction method that takes semantic features and prior information of strokes into consideration. This method consists of three parts: image registration-based stroke registration that establishes the rough registration of the reference strokes and the target as prior information; image semantic segmentation-based stroke segmentation that preliminarily separates target strokes into seven categories; and high-precision extraction of single strokes. In the stroke registration, we propose a structure deformable image registration network to achieve structure-deformable transformation while maintaining the stable morphology of single strokes for character images with complex structures. In order to verify the effectiveness of the method, we construct two datasets respectively for calligraphy characters and regular handwriting characters. The experimental results show that our method strongly outperforms the baselines. Code is available at https://github.com/MengLi-l1/StrokeExtraction.